1. Field of the Invention
The present invention relates to a performance predictive apparatus, a method, and a recording medium, in particular, suitably used for a system for evaluating the performance of a computer system built with machines, a network, etc.
2. Description of the Related Art
Generally, in case of building a certain computer system, the system is built through processes including proposal of system contents, definition of requirements, design of details, implementation, and testing, and then put into operation. Such system building is normally performed in accordance with the needs or requirement (e.g., the request within how many seconds the data read on the database should be restrained) of a customer or system user. To meet the needs or requirement, as the server machine, the client machine, and the network for connecting them, to build the system, appropriate ones must be used.
Conventionally, however, system sizing such as that what specifications of server and clients should be put in the computer system to be built or that what capacity of network should be introduced, depends on experience or intuition of the system developer. Whether the customer's needs or requirement are met, is evaluated by actually operating the system at the time of completing the system. In case of changing the network topology of the system wherein various types of hardware are connected via a network, the problem of what distribution of loads on the system is also dependent on experience or the like.
For this reason, the system thus built depending on experience or intuition, often results in customer's undesirability. In this case, there may arise the necessity of rebuilding the system to meet the needs or requirement of the customer, and there is the problem of resulting in requiring large labor and high cost.
So, in recent years, to avoid such an inconvenience, a simulation model of the system to be built is first created on a computer, and simulation is executed by inputting various parameters to the model, and thereby the system performance is estimated. On the basis of the estimation result, the system configuration or sizing of each machine is performed.
But, even in case of using such simulation, the parameters to be input to the simulation model are determined by the system developer on the basis of his or her own experience or intuition. For this reason, there is a difference in estimation result of performance according to the person in charge, and, in some cases, the system may be built on the basis of an erroneous performance estimation result. In this case, there arises the same problem as described above.
In addition, because, in the conventional simulator, an absolute value (e.g., an absolute time of how many seconds are required for performing a certain process by using a CPU of a certain specification) predicted by the developer or the like, is input, even when the specification of a machine included in the simulation model are changed, the response time is unchangeable as an absolute value, and the obtained performance estimation result becomes the same, and there is the problem of lacking in reliability.
Besides, particularly at the early stage of system development before an order or before long after the order, only an outline of the system to be built is known, and its minute construction is not yet made to be detailed. So, at this stage, a detailed simulation model is hard to create, and it cannot but be only an approximate model, and the amount of information necessary for parameter input is also small. For this reason, there are also the problems that it is difficult to input appropriate parameters, and it is very hard to evaluate the system performance accurately.